• Annual Conference of KIPS
• /
• 2023.11a
• /
• pp.672-674
• /
• 2023

본 논문은 뉴로-심볼릭 구조를 바탕으로 일반 텍스트로부터 온톨로지 생성이 가능한 심층 신경망 기반 온톨로지 추출기를 제안한다. 온톨로지 추출 단계를 (i) 온톨로지 학습 및 (ii) 온톨로지 생성의 2 단계로 상정, (i) 일반 텍스트로부터 문장 구조 및 논리적 관계를 학습하는 트랜스포머 기반 심층 생성 신경망 출력을 이용하여 (ii) 계층적으로 결합한 심볼릭 추론기로 온톨로지를 생성하는 뉴로-심볼릭 구조 온톨로지 추출기를 구현하였다. 1800 개 훈련 집합으로 학습 후 200 개 테스트 집합으로 평가한 결과, 정확도 91.9%, Precision 100%, Recall 99.1%로 비교 모델 OpenIE 의 성능에 비해서 각각 83.8%, 1.8%, 3.5% 개선된 것을 확인하였다. 정성적 품질에 있어서, 복잡한 문장 (예: 관계대명사, 접속사, 중첩 구조)에서도 비교 모델에 비해 더 정밀한 온톨로지 생성 결과를 보였다.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.1
• /
• pp.253-262
• /
• 2016

This paper proposes a balance-swap method for the dynamic economic load dispatch problem. Based on the premise that all generators shall be operated at valve-points, the proposed algorithm initially sets the maximum generation power at $P_i{\leftarrow}P_i^{max}$. As for generator i with $_{max}c_i$, which is the maximum operating cost $c_i=\frac{F(P_i)-F(P_{iv_k})}{(P_i-P_{iv_k})}$ produced when the generation power of each generator is reduced to the valve-point $v_k$, the algorithm reduces i's generation power down to $P_{iv_k}$, the valve-point operating cost. When ${\Sigma}P_i-P_d$ > 0, it reduces the generation power of a generator with $_{max}c_i$ of $c_i=F(P_i)-F(P_i-1)$ to $P_i{\leftarrow}P_i-1$ so as to restore the equilibrium ${\Sigma}P_i=P_d$. The algorithm subsequently optimizes by employing an adult-step method in which power in the range of $_{min}\{_{max}(P_i-P_i^{min}),\;_{max}(P_i^{max}-P_i)\}$>${\alpha}{\geq}10$ is reduced by 10; a baby step method in which power in the range of 10>${\alpha}{\geq}1$ is reduced by 1; and a swap method for $_{max}[F(P_i)-F(P_i-{\alpha})]$>$_{min}[F(P_j+{\alpha})-F(P_j)]$, $i{\neq}j$ of $P_i=P_i{\pm}{\alpha}$, in which power is swapped to $P_i=P_i-{\alpha}$, $P_j=P_j+{\alpha}$. It finally executes minute swap process for ${\alpha}=\text{0.1, 0.01, 0.001, 0.0001}$. When applied to various experimental cases of the dynamic economic load dispatch problems, the proposed algorithm has proved to maximize economic benefits by significantly reducing the optimal operating cost of the extant Heuristic algorithm.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.2
• /
• pp.165-182
• /
• 2001

In this paper, we describe a design and implementation of the Wireless Markup Language(WML) compiler to deploy wireless Internet services effectively. The WML compiler translates textual WML decks into binary ones in order to reduce the traffic on wireless links that have relatively low bandwidth to wireline links and mitigate the processing overhead of WML decks on, wireless terminals that have relatively low processing power to fixed workstations. In addition, it takes over the overhead of eXtensible Markup Language(XML) well-formedness and validation processes. The WML compiler consists of the lexical analyzer and parser modules. The granunar for the WML parser module is LALR(1) context-free grammar that is designed based on XML 1.0 and WML 1.2 DTD(Document Type Definition) with the consideration of the Wireless Application Protocol Binary XML grammar. The grammar description is converted into a C program to parse that grammar by using parser generator. Even though the tags in WML will be extended or WML DTD will be upgraded, this approach has the advantage of flexibility because the program is generated by modifying just the changed parts. We have verified the functionality of the WML compiler by using a WML decompiler in the public domain and by using the Nokia WAP Toolkit as a WAP client. To measurethe compressibility gain of the WML compiler, we have tested a large number of textual WML decks and obtained a maximum 85 %. As the effect of compression is reduced when the portion of general textual strings increases relative to one of the tags and attributes in a WML deck, an extended encoding method might be needed for specific applications such as compiling of the WML decks to which the Hyper Text Markup Language document is translated dynamically.

• Journal of Service Research and Studies
• /
• v.14 no.1
• /
• pp.13-26
• /
• 2024

Multi-modal generation is the process of generating results based on a variety of information, such as text, images, and audio. With the rapid development of AI technology, there is a growing number of multi-modal based systems that synthesize different types of data to produce results. In this paper, we present an AI system that uses speech and text recognition to describe a person and generate a montage image. While the existing montage generation technology is based on the appearance of Westerners, the montage generation system developed in this paper learns a model based on Korean facial features. Therefore, it is possible to create more accurate and effective Korean montage images based on multi-modal voice and text specific to Korean. Since the developed montage generation app can be utilized as a draft montage, it can dramatically reduce the manual labor of existing montage production personnel. For this purpose, we utilized persona-based virtual person montage data provided by the AI-Hub of the National Information Society Agency. AI-Hub is an AI integration platform aimed at providing a one-stop service by building artificial intelligence learning data necessary for the development of AI technology and services. The image generation system was implemented using VQGAN, a deep learning model used to generate high-resolution images, and the KoDALLE model, a Korean-based image generation model. It can be confirmed that the learned AI model creates a montage image of a face that is very similar to what was described using voice and text. To verify the practicality of the developed montage generation app, 10 testers used it and more than 70% responded that they were satisfied. The montage generator can be used in various fields, such as criminal detection, to describe and image facial features.

• Journal of Korea Multimedia Society
• /
• v.6 no.6
• /
• pp.1088-1099
• /
• 2003

The syntax-directed translation using semantic actions is frequently used in construction of compiler for text programming languages. it is very useful for the language designers to develop compiler back-end using a syntax structure of a source programming language. Due to the lack of the integrated representation method for a parse tree node and modeling method of syntax structures, it is very hard to construct compiler using syntax-directed translation in visual languages. In this Paper, we propose a visual language compiler generation method for constructing a visual languages compiler automatically, using syntax-directed translation. Our method uses the Picture Layout Grammar as a underlying grammar formalism. This grammar allows our approach to generate parser efficiently u sing And-Or-Waiting Graph and encapsulating syntax definition as one unit. Unlike other systems, we suggest separating the specification and the generation of semantic actions. Because of this, it provides a very efficient method for modification.

• Progress in Medical Physics
• /
• v.23 no.4
• /
• pp.303-308
• /
• 2012

The Geant4 based Monte Carlo code for the application of stereotactic radiosurgery was developed. The probability density function and cumulative density function to determine the incident photon energy were calculated from pre-calculated energy spectrum for the linac by multiplying the weighting factors corresponding to the energy bins. The messenger class to transfer the various MLC fields generated by the planning system was used. The rotation matrix of rotateX and rotateY were used for simulating gantry and table rotation respectively. We construct accelerator world and phantom world in the main world coordinate to rotate accelerator and phantom world independently. We used dicomHandler class object to convert from the dicom binary file to the text file which contains the matrix number, pixel size, pixel's HU, bit size, padding value and high bits order. We reconstruct this class object to work fine. We also reconstruct the PrimaryGeneratorAction class to speed up the calculation time. because of the huge calculation time we discard search process of the ThitsMap and used direct access method from the first to the last element to produce the result files.

• Journal of the Institute of Convergence Signal Processing
• /
• v.23 no.4
• /
• pp.222-227
• /
• 2022

In this paper, we implemented a Korean text generation and classification model based on a deep learning algorithm that can be applied to various industries. It consists of two implemented GAN-based Korean handwriting generation models and CNN-based Korean handwriting classification models. The GAN model consists of a generator model for generating fake Korean handwriting data and a discriminator model for discriminating fake handwritten data. In the case of the CNN model, the model was trained using the 'PHD08' dataset, and the learning result was 92.45. It was confirmed that Korean handwriting was classified with % accuracy. As a result of evaluating the performance of the classification model by integrating the Korean cursive data generated through the implemented GAN model and the training dataset of the existing CNN model, it was confirmed that the classification performance was 96.86%, which was superior to the existing classification performance.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.5
• /
• pp.803-808
• /
• 2024

As a service that creates graphic work images with AI, DALL-E2, Midjourney, Stable Diffusion, BING image generator, and Playground AI are widely used. It is that graphic also enables learner-led customized education. With this, it is worth studying detailed design customized learning materials and methods for designing efficient design in future 2D graphic work, and it is necessary to explore the areas of application. The current situation is that it is necessary to develop a design education system that can indicate the lack of AI technology through text security and questions. In this study, a successful proposal for a process that is produced through a process of creating AI design work through proxy work can be presented as a conclusion. Design, advertisement, and visual content companies are already using and adapting, and the trend is to reflect the AI graphic utilization ability and results in the portfolio along with interviews when hiring new employees. In line with this, detailed consideration and research on visual and design production methods for AI convergence between instructors and learners are currently needed. In this paper, proposals and methods for image quality production were considered in the main body and conclusions, and conclusive directions were proposed for five alternatives and methods for future applications.